Institut de Lingüística Aplicada

Third Mediterranean Meeting on Morphology (MMM3)


Sergei Tatevosov (Moscow State University)
Verbal reduplication in Bagwalal and Fixed Segmentism

The present paper aims at providing an account for reduplication in Bagwalal (Andic, East Caucasian), and discussing theoretical issues this account touches upon within the broader context of Optimality Theory (Prince & Smolensky 1993).

Problem. It is commonly assumed within OT framework that inexact copying of the base by the reduplicant must be attributed to the general ranking schema in which markedness constraints dominate Base-Reduplicant faithfulness constraints, being in turn dominated by Input-Output faithfulness constraints (This is referred to as the emergence of the unmarked, see McCarthy & Prince 1994, 1995, Alderete et al. 1998, among others). In Bagwalal there are at least two constraints outranking B-R identity. In the present paper I will (a) identify type of these constrains; (b) establish ranking for the constraints; (c) discuss implications for the factorial typology that follow from this ranking. Also, I will test predictions made by Correspondence Theory (McCarthy, Prince 1995) against those derived from recently developed Theory of Morphological Doubling (Inkelas, Zoll 2000).

Data and informal analysis. Data in (1) suggest that the domain for reduplication is a root which is maximally bisyllabic, with final monomoraic light syllable. I assume -ri to be a derivational affix forming a morphological constituent called Derivational Stem, as represented in fig.1:

  1. Input Output Gloss
    /zàoq'u/ zào [q'o]R[q'u]B 'get tired'
    /kwaXi/ kwa [Xa]R[Xi]B 'steal'
    /biha)/ bi[hi]R[ha)] 'boil, cook'
    /qasiri/ qa[sa]R[si]Bri 'scratch'
    /k'anc'uri/ k'an[c'a]R[c'u]Bri 'jump'
    /Runkiri/ Run[ku]R[ki]Bri 'unite'
    [+ tree]

In Bagwalal reduplication, the base is a final light syllable of the root, and the reduplicant is prefixed to it, resulting pattern being infixation.

The main phenomenon to be accounted for is that a vowel in B is never copied by R: it is either identical to the vowel in the initial syllable of the root, as in (1), or realized as a default segment /i/, as in (2):

  1. q'ocàa / q'o[cài]R[càa]B / 'love'
    baqwa / ba[qu]R[qwa]B / 'dig' (/i/ surfaces as /u/ after labialized consonants)

To account for (1), I follow long-standing tradition of treating R as an affix (McCarthy 1979, 1981; Maranz 1982) that lack any specific segmental content; R stands in correspondence relation with B, being subject to B-R identity constraints (McCarthy & Prince 1995). The fact that copying of the vowel fails in R implies that there is another constraint C, outranking B-R identity. I assume that C is a feature alignment constraint that requires features linked to the vowel in the initial syllable to spread to the right edge of the relevant domain. Spreading is blocked if the full feature specification is prelinked to a vowel in the input. I assume, then, that every segment in the input has full feature specification, and, as a result, C can only take effect on R.

(2) is an instance of the fixed segmentism. 'Default' status of /i/ is motivated independently by the fact that it is epenthesized to resolve unsyllabifiable sequences occurring within loan words (isàkola || *sàkola 'school') or arising due to affixation (/misuX+s/: misuXis 'house; adelative'). Fixed segment /i/ figures in R to satisfy the following co-occurrence restriction: [+low] in B must correspond to [+high] in R. This restriction, which is evidently a manifestation of Obligatory Contour Principle, is ranked higher than alignment constraint that requires spreading attested in (1), so it is better to insert /i/ than to create the output violating OCP (*baqaqwa; q'ocà'ocà'a).

To see the fullest spectrum of reduplication possibilities consider (3):

  1. eXi / e[Xi]R[Xi]B / 'hang'
    hec'i / he[c'i]R[c'i]B / 'rise'

It follows from the above discussion that in (3) /i/ in R is not motivated by B-R identity, but should be treated as another instance of fixed segmentism that emerges as an effect of some constraint outranking feature alignment constraint responsible for (1). I suggest that here we are dealing with a kind of morpheme shape condition saying that no affix in Bagwalal is allowed to contain the segment /e/. As R is an affix, spreading features from the initial syllable will result in a configuration that violates this condition.